OpenClaw MCP Server

Model Context Protocol (MCP) server for Heretek OpenClaw skills and memory exposure

Overview

The OpenClaw MCP Server provides standardized access to Heretek OpenClaw capabilities through the Model Context Protocol. It exposes:

• Resources: Agent memories, knowledge base, skill definitions
• Tools: Skill execution endpoints, memory operations, knowledge search
• Prompts: Common agent interaction templates

This implementation supersedes the basic MCP client in openclaw-mcp-connectors by providing a full MCP server that external clients can connect to.

Installation

npm install @heretek-ai/openclaw-mcp-server

Usage

As a Standalone Server

# Using npx
npx openclaw-mcp-server

# Or directly
node src/index.js

With Environment Variables

OPENCLAW_SKILLS_PATH=./skills \
OPENCLAW_MEMORY_PATH=./memory \
OPENCLAW_KNOWLEDGE_PATH=./knowledge \
npx openclaw-mcp-server

Configuration

| Environment Variable | Default | Description | |---------------------|---------|-------------| | OPENCLAW_SKILLS_PATH | ./skills | Path to OpenClaw skills directory | | OPENCLAW_MEMORY_PATH | ./memory | Path to memory storage directory | | OPENCLAW_KNOWLEDGE_PATH | ./knowledge | Path to knowledge base directory |

Exposed Resources

Memory Resources

| URI | Description | |-----|-------------| | memory://episodic/list | List all episodic memories | | memory://episodic/{id} | Get specific episodic memory | | memory://semantic/list | List all semantic schemas | | memory://semantic/{schemaId} | Get specific semantic schema | | memory://session/list | List all agent sessions | | memory://session/{agentId} | Get agent session memory | | memory://swarm/stats | Get swarm memory statistics |

Knowledge Resources

| URI | Description | |-----|-------------| | knowledge://docs/list | List all documents | | knowledge://docs/{path} | Get specific document | | knowledge://schemas/list | List knowledge schemas | | knowledge://schemas/{id} | Get specific schema | | knowledge://graph/stats | Get knowledge graph statistics | | knowledge://ingest/status | Get ingestion status |

Skill Resources

| URI | Description | |-----|-------------| | skill://list | List all available skills | | skill://{name} | Get specific skill definition | | skill://categories | List skill categories | | skill://category/{category} | List skills in category |

Exposed Tools

Skill Tools

| Tool | Description | |------|-------------| | skill-execute | Execute any OpenClaw skill by name | | skill-list | List all available skills | | skill-info | Get information about a skill | | skill-{name} | Quick access to specific skills |

Memory Tools

| Tool | Description | |------|-------------| | memory-search | Search across memory using natural language | | memory-read | Read specific memory by ID | | memory-stats | Get swarm memory statistics | | memory-consolidate | Trigger memory consolidation |

Knowledge Tools

| Tool | Description | |------|-------------| | knowledge-search | Hybrid search (vector + keyword) | | knowledge-read | Read specific document | | knowledge-ingest | Ingest new document | | knowledge-graph-query | Query knowledge graph |

Exposed Prompts

| Prompt | Description | |--------|-------------| | agent-deliberation | Triad deliberation template | | agent-proposal | Create new proposal | | agent-safety-review | Sentinel safety review | | agent-memory-query | Memory query template | | agent-knowledge-search | Knowledge search template | | agent-skill-execution | Skill execution request | | agent-explorer-intel | Explorer intelligence request | | agent-historian-retrieval | Historian retrieval request | | agent-coder-implementation | Coder implementation request | | agent-dreamer-synthesis | Dreamer synthesis request | | agent-empath-user-context | Empath user context query | | agent-steward-orchestrate | Steward orchestration request |

Integration with MCP Connectors

The MCP Server works alongside the existing openclaw-mcp-connectors plugin:

┌─────────────────────────────────────────────────────────────────┐
│                    MCP Integration Architecture                  │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  ┌──────────────────────┐         ┌──────────────────────┐     │
│  │  External MCP        │         │  OpenClaw MCP        │     │
│  │  Clients             │         │  Connectors          │     │
│  │  (IDEs, Agents)      │         │  (Client-side)       │     │
│  └──────────┬───────────┘         └──────────┬───────────┘     │
│             │                                │                  │
│             │  MCP Protocol                  │  MCP Protocol    │
│             │  (stdio/sse)                   │  (stdio/sse)     │
│             ▼                                ▼                  │
│  ┌──────────────────────────────────────────────────────────┐   │
│  │              OpenClaw MCP Server                          │   │
│  │                                                           │   │
│  │  ┌─────────────┐  ┌─────────────┐  ┌─────────────────┐   │   │
│  │  │  Resources  │  │   Tools     │  │    Prompts      │   │   │
│  │  │  Handler    │  │  Handler    │  │    Handler      │   │   │
│  │  └─────────────┘  └─────────────┘  └─────────────────┘   │   │
│  └──────────────────────────────────────────────────────────┘   │
│             │                                │                  │
│             ▼                                ▼                  │
│  ┌──────────────────────┐         ┌──────────────────────┐     │
│  │  OpenClaw Skills     │         │  OpenClaw Memory     │     │
│  │  (48 skills)         │         │  (3-tier system)     │     │
│  └──────────────────────┘         └──────────────────────┘     │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘

Connection Example

// External client connecting to OpenClaw MCP Server
const { Client } = require('@modelcontextprotocol/sdk/client/index.js');
const { StdioClientTransport } = require('@modelcontextprotocol/sdk/client/stdio.js');

const transport = new StdioClientTransport({
  command: 'npx',
  args: ['openclaw-mcp-server'],
});

const client = new Client({
  name: 'my-agent',
  version: '1.0.0',
});

await client.connect(transport);

// List available resources
const resources = await client.request({ method: 'resources/list' });

// Read a resource
const memory = await client.request({
  method: 'resources/read',
  params: { uri: 'memory://episodic/list' },
});

// Execute a skill
const result = await client.request({
  method: 'tools/call',
  params: {
    name: 'skill-execute',
    arguments: {
      skillName: 'healthcheck',
      arguments: [],
    },
  },
});

// Get a prompt
const prompt = await client.request({
  method: 'prompts/get',
  params: {
    name: 'agent-deliberation',
    arguments: {
      proposal: 'Implement new feature X',
      priority: 'high',
    },
  },
});

Architecture

Handler Structure

src/
├── index.js                 # Main server entry point
└── handlers/
    ├── memory-resources.js  # Memory resource handler
    ├── knowledge-resources.js # Knowledge resource handler
    ├── skill-resources.js   # Skill resource handler
    ├── skill-tools.js       # Skill tool handler
    └── prompts.js           # Prompt template handler

Data Flow

1. MCP Client Request
         │
         ▼
2. MCP Server (index.js)
         │
         ▼
3. Request Handler Routing
   ├── Resources → Memory/Knowledge/Skill handlers
   ├── Tools → Skill/Memory/Knowledge tool handlers
   └── Prompts → Prompt handler
         │
         ▼
4. OpenClaw Resources
   ├── Skills (48 skills)
   ├── Memory (3-tier system)
   └── Knowledge (documents, schemas, graph)
         │
         ▼
5. Response formatted as MCP result

Development

Testing

npm test

Linting

npm run lint

Adding New Resources

1. Create handler in src/handlers/
2. Implement listResources() and readResource(uri)
3. Register in main server index.js

Adding New Tools

1. Implement tool in appropriate handler
2. Add to listTools() return array
3. Add handler in callTool(name, args)
4. Register in main server index.js

Adding New Prompts

1. Add template definition in prompts.js
2. Implement _generate{PromptName}Prompt() method
3. Add case in _generatePrompt() switch

License

MIT

References

• Model Context Protocol Specification
• OpenClaw MCP Connectors
• OpenClaw Skills Documentation
• Swarm Memory Architecture